Stronger Synaptic Connectivity As a Mechanism Behind Development of Working Memory–Related Brain Activity During Childhood

Stronger Synaptic Connectivity as a Mechanism behind Development of Working Memory–related Brain Activity during Childhood

Fredrik Edin1,2, Julian Macoveanu1, Pernille Olesen1, Jesper Tegne´r1,3,4,

and Torkel Klingberg1 Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/5/750/1756680/jocn.2007.19.5.750.pdf by guest on 18 May 2021

Abstract & The cellular maturational processes behind cognitive de- compared to brain activity measured with functional magnetic velopment during childhood, including the development of resonance imaging in children and adults. We found that net- working memory capacity, are still unknown. By using the works with stronger fronto-parietal synaptic connectivity be- most standard computational model of visuospatial working tween cells coding for similar stimuli, but not those with faster memory, we investigated the consequences of cellular matura- conduction, stronger connectivity within a region, or increased tional processes, including myelination, synaptic strengthen- coding specificity, predict measured developmental increases ing, and synaptic pruning, on working memory-related brain in both working memory-related brain activity and in correla- activity and performance. We implemented five structural de- tions of activity between regions. Stronger fronto-parietal syn- velopmental changes occurring as a result of the cellular matu- aptic connectivity between cells coding for similar stimuli was rational processes in the biophysically based computational thus the only developmental process that accounted for the network model. The developmental changes in memory activ- observed changes in brain activity associated with develop- ity predicted from the simulations of the model were then ment of working memory during childhood. &

INTRODUCTION Klingberg, Forssberg, & Westerberg, 2002; Casey, Giedd, & Working memory (WM), the ability to temporarily main- Thomas, 2000; Giedd et al., 1999; Sowell, Thompson, tain visuospatial information in mind, is a key cognitive Holmes, Jernigan, & Toga, 1999). Despite the fact that function that underlies other cognitive abilities such as interpretations of the developmental changes in brain complex reasoning, and undergoes significant maturation activity have been made by referring to structural matura- during childhood and adolescence (Gathercole, Pickering, tional processes (Bunge et al., 2002; Casey et al., 2000), Ambridge, & Wearing, 2004; Westerberg, Hirvikoski, it has actually not yet been demonstrated whether the Forssberg, & Klingberg, 2004; Fry & Hale, 2000; Baddeley maturational changes occurring at the cellular level really & Hitch, 1974). Several maturational processes take place result in the gross changes in brain activity associated during that time, most importantly, the myelination of with cognitive development, nor has it been demonstrated axons, the strengthening of synapses, and synaptic prun- how this process may occur. For example, how would ing (Bourgeois, Goldman-Rakic, & Rakic, 2000; Lamantia myelination, which increases signal conduction velocity, & Rakic, 1990; Rakic, Bourgeois, Eckenhoff, Zecevic, affect macroscopic brain activity as measured with fMRI? & Goldman-Rakic, 1986; Huttenlocher, 1979; Yakovlev & Does synaptic pruning cause increases or decreases in Lecours, 1967; Hubel & Wiesel, 1963). Anatomical and brain activity? Here, by integrating computational model- functional magnetic resonance imaging (fMRI) has ing, with which we can predict the effects of structural been used to map structural and physiological changes changes on brain activity, and fMRI, with which we can associated with cognitive development during child- learn which of the predicted changes in brain activity ac- hood (Olesen, Nagy, Westerberg, & Klingberg, 2003; tually occur, we make the connection between structural Bunge, Dudukovic, Thomason, Vaidya, & Gabrieli, 2002; changes and physiological development during the maturation of visuospatial WM (vsWM) in the adolescent. The computational model that we use to make pre- 1Karolinska Institutet, Stockholm, Sweden, 2Royal Institute of dictions about the development of macroscopic brain Technology, Stockholm, Sweden, 3Stockholm Bioinformatics activity underlying vsWM relies on knowledge about the Center, Stockholm, Sweden, 4Linko¨ping University of Tech- neuronal basis of vsWM. Specifically, electrophysiologi- nology, Sweden cal experiments on behaving monkeys reveal sustained

D 2007 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 19:5, pp. 750–760 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.2007.19.5.750 by guest on 30 September 2021 neuronal activity during the delay period of vsWM trials were put forth, each describing in detail a single aspect (Funahashi, Bruce, & Goldman-Rakic, 1989). The activity of structural development. Thus, the effect of each struc- is cue-specific so that different neurons code for objects tural change could be studied in isolation. The develop- at different angles in the visual field. In parallel with the mental change of each hypothesis could be modeled by characterization of vsWM-related activity on the neuro- a simple change of a single parameter in the network nal level, progress in basic cortical physiology has pro- model. For each hypothesis, we produced a ‘‘child’’ ver- duced a detailed description of cellular and synaptic sion and an ‘‘adult’’ version of the network, differing only characteristics of pyramidal cells and inhibitory inter- in the parameter change relating to that specific hypothe- neurons (Douglas & Martin, 2004). These findings were sis. The ‘‘adult’’ version of the network was common to Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/5/750/1756680/jocn.2007.19.5.750.pdf by guest on 18 May 2021 recently incorporated in a biophysically based computa- all hypotheses. Simulations were performed with each of tional network model of vsWM activity (Wang, Tegne´r, the networks, and the difference between characteristics Constantinidis, & Goldman-Rakic, 2004; Tegne´r, Compte, of the simulated BOLD activity in the ‘‘child’’ and ‘‘adult’’ & Wang, 2002; Wang, 2001; Compte, Brunel, Goldman- networks of each hypothesis was the prediction from Rakic, & Wang, 2000; Amit & Brunel, 1997), from which that hypothesis. In order to test the prediction of the the model in this study was developed. The model has model, we let a group of children and a group of adults been able to explain several characteristics of the activ- perform a vsWM task while the delay-phase BOLD activity ity in the frontal cortex (Funahashi et al., 1989) during was measured. In this way, we could conclude which of the performance of a WM task, as reviewed by Compte the hypotheses could accurately predict experimentally (2006). It reproduces a low but stable activity during measured developmental changes in brain activity relat- the fixation period, as well as a higher and stable mem- ing to vsWM and which could not. ory activity with physiological firing rates during the delay period. It also explains the decrease in delay- period activity in cells not coding for the presented METHODS visual stimulus. Furthermore, it has predicted differential connection strength between neurons depend- To make the logical structure of the study easier to ing on the degree of similarity of the stimuli that they grasp, the more technical parts concerning modeling encode. This has later been confirmed in experiments and fMRI data acquisition have been moved to the (Constantinidis, Franowicz, & Goldman-Rakic, 2001), Supplemental Methods section. indicating some predictive power of the model. Unfortunately, the original version of the model only Computational Model: General Overview describes vsWM activity in one region of the frontal cortex, whereas vsWM studies in humans (Curtis, Rao, The structure of the vsWM network model that we creat- & D’Esposito, 2004; Rowe, Toni, Josephs, Frackowiak, & ed is shown in Figure 1A and B. The network contains Passingham, 2000; Courtney, Petit, Maisog, Ungerleider, two interconnected regions, each consisting of a popu- & Haxby, 1998) and monkeys (Chafee & Goldman- lation of 128 pyramidal cells (P) and a population of Rakic, 1998, 2000) have found sustained delay activity 32 inhibitory interneurons (I). Every cell codes for an associated with vsWM in several regions, most consist- angle in the visual field. The two regions are replicas of ently in the superior frontal sulcus (SFS, presumably the frontal region network in Tegne´r et al. (2002), and monkey area 8a) and the intraparietal sulcus (IPS, like that model, this model also consists of Hodgkin– presumably monkey area 7ip). Therefore, in order to Huxley type cells with ion channels and input–output evaluate hypotheses about which neuronal developmen- relations matching those of real layer II/III neurons. The tal process accounts for the developmental improve- regions are connected only through their pyramidal ment in vsWM, we extended the single-region vsWM cells. Interregional connections have a conduction delay computational network model to a two-region model. (Ferraina, Pare, & Wurtz, 2002), whereas all other con- This allowed us to investigate hypotheses concerning nections are instantaneous. There exists a topography in strengthening of synapses, both within a region and be- the connection strength between pyramidal cells within tween regions, as well as synaptic pruning and the ve- or between two regions, as indicated by the connection locity of signal conduction between regions. To compare curve (Figure 1B), a key term in this text as several of the results of the model to experimental results, we the developmental hypotheses are expressed as changes translated the neuronal activity of the model into blood in this curve. The curve shows that cells with similar oxygenation level-dependent (BOLD) signals (Deco, preferred angles are strongly connected, whereas cells Rolls, & Horwitz, 2004), the signals measured with fMRI. with dissimilar preferred angles are weakly connected The organization of the investigations aiming to con- (Constantinidis et al., 2001). nect the structural and functional development of vsWM Simulations (Figure 1C) showed that the model ac- was as follows: From the general observation that syn- counts for several characteristics of the delay-phase neu- aptic pruning, synaptic strengthening, and myelination ral activity in the vsWM tasks (Chafee & Goldman-Rakic, take place during development, five specific hypotheses 1998; Funahashi et al., 1989). As in previous, single-region

Edin et al. 751 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.2007.19.5.750 by guest on 30 September 2021 versions of the model, model neurons showed the ex- activity of the network into a corresponding BOLD sig- perimentally observed stable resting activity during the nal (Deco et al., 2004; Figure 1D). Total synaptic current intertrial interval as well as stable, spatially localized and (Attwell & Iadecola, 2002), as given by the sum of the ab- physiologically realistic mnemonic firing rates during the solute values of the AMPA (a-amino-3-hydroxy-5-methyl- delay phase. The model also reproduced the decrease in 4-isoxazoleproprionic acid), NMDA (N-methyl-D-aspartic activity in cells not coding for the memory during the acid), and GABA (g-aminobutyric acid) components, with delay phase. The activity in the two simulated regions a standard hemodynamic response function (Friston was very similar, which is in agreement with single-unit et al., 1998). We confirmed that the use of spiking activity recordings from the frontal and parietal cortices in the (Mukamel et al., 2005) or only excitatory synaptic activity Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/5/750/1756680/jocn.2007.19.5.750.pdf by guest on 18 May 2021 macaque (Chafee & Goldman-Rakic, 1998). as a basis for the BOLD signal did not affect our results. To be able to compare simulations from the model The neural basis of the BOLD signal has been discussed with experimental results, we translated the electrical by Attwell and Iadecola (2002).

Developmental Hypotheses Having established the computational model with which we can predict the effect of our developmental hypotheses on the BOLD signal, we are now in a position to describe these. Based on the three known neuronal developmental changes, we put forth a total of five hypotheses (Figure 2A). The first two hypotheses are exclusively related to strengthening of synapses whereas the following two are related to both synaptic strengthening and pruning. All four are expressed as changes in the neuronal connection curve. Finally, the fifth hypothesis is related to myelination. The five hypotheses will now be described in order. The first hypothesis (H1) modeled a developmental strengthening of connections within a region by increasing the mean frontal and parietal intraregional connection strengths Gff and Gpp (terms in this subsection are explained in Figure 1B and the Supplementary Methods section). The second hypothesis (H2) tested a strength-

Figure 1. The two-region network. (A) The vsWM network with a frontal cortical (FC) and a parietal cortical (PC) region. Each region consists of a pyramidal cell population (P) and an inhibitory interneuron (I) population, both connected internally and with the other population. Interregional connections are purely excitatory. The gray curves are connection curves (see B). (B) The connection curve indicates how the connection strength between two pyramidal cells within or between two regions depends on the difference in their preferred angle. In the model, the connection curve has the

shape of a Gaussian curve on top of a box. Gxy is the mean connection + strength from area y onto area x. Gxy J is the height of the connection curve, and s is the standard deviation of the Gaussian curve. To regulate the shape of the connection curve while preserving total connection strength (area under curve), changes in s or J+ are compensated by changes in J. (C) Network simulation of the WM trial. Dots represent action potentials. Pyramidal cells are aligned according to stimulus specificity. Cue stimuli enter both regions as a current into pyramidal cells coding for the stimulus. Memory for the position (08 u < 3608) of a visual cue is retained during the delay phase through localized persistent activity. At selection, a current causes memory activity to return to the baseline. (D) By convolving the network synaptic currents (thin) with a standard hemodynamic response (HR) function (Friston et al., 1998, inset), a simulated BOLD signal was obtained (Deco et al., 2004, thick). Red = pyramidal cells; blue = inhibitory interneurons.

752 Journal of Cognitive Neuroscience Volume 19, Number 5 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.2007.19.5.750 by guest on 30 September 2021 Figure 2. The logical structure of the study. (A) Simulations. Based on three known neuronal developmental processes (column 1), we put forth five hypotheses (H1–H5) regarding the structural development of the vsWM network (column 2). For each hypothesis, ‘‘child’’

(black) and ‘‘adult’’ (green) Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/5/750/1756680/jocn.2007.19.5.750.pdf by guest on 18 May 2021 versions of the network were created. The strengths of connections within a region are indicated by the connection curves inside the circles (which represent the parietal and frontal pyramidal cell populations), whereas the curves between the circles show connection curves between regions. Finally, predictions from the hypotheses regarding the development of the vsWM maintenance-related BOLD signal were obtained through simulation (column 3). H1: greater connection strength within regions. H2: greater connection strength between regions. H3: higher contrast. H4: higher specificity. H5: faster signal conduction velocity (indicated by the arrows). In the hypotheses, structural development occurs at connections within (w) and/or between (b) regions. (B) Experiments. To find which of the developmental processes that can predict the development of vsWM maintenance-related BOLD signals, fMRI data were collected from children and adults during the performance of a vsWM task. Column 2 shows pooled adult and child delay-phase activity. Consistent with the model, frontal and parietal regions (encircled) were significantly activated. Column 3 shows the delay-phase BOLD signal in each group separately, averaged across bilateral SFS and IPS regions (arbitrary units).

ening of connections between two regions by increas- neuron provides more specific information about the ing the mean interregional connection strengths Gfp identity of the stimulus. An early example of these and Gpf. phenomena is Hubel and Wiesel’s study showing how Hypotheses 3 and 4 (H3, H4) modeled development cells in the visual cortex first receive widespread inputs, as synaptic remodeling leading to a redistribution of but that elimination results in neuronal responses hav- synapses without changing total connection strength. ing higher contrast and greater specificity (Hubel & This could occur through the simultaneous pruning of Wiesel, 1963). Later, Rainer and Miller (2000) found silent synapses and strengthening of active synapses, so increased specificity in the monkey prefrontal cortex as H3 and H4 are related to both synaptic strengthening a result of visual training. Thus, the third hypothesis and pruning. Two changes are possible: increasing J+ (H3) represented a higher contrast by increasing J+, and leads to a higher contrast in the neuronal response, and the fourth hypothesis (H4) represented increased neu- decreasing s leads to a sharper connection curve with ronal specificity by decreasing s. higher neuronal coding specificity. An increased contrast The third developmental process, myelination, con- means that the difference in activity between neurons tinues throughout adolescence (Yakovlev & Lecours, coding for the stimulus and neurons with unrelated 1967) and is usually thought to improve cognitive pro- activity will be greater. An increased specificity means cessing by increasing action potential conduction veloc- that, in order to activate the neuron, stimuli must be ity. Therefore, to model the effect of myelination, we more similar to the preferred stimulus of that neuron designed a model (H5) with increased action conduc- than was previously the case; hence, the activity of the tion velocity between the two cortical regions. To test

Edin et al. 753 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.2007.19.5.750 by guest on 30 September 2021 the hypotheses, we designed one ‘‘adult’’ and five one of the previously presented dots for 0.5 sec, and ‘‘child’’ networks (Figure 2A, column 2). The ‘‘child’’ subjects had to click on the intersection between the networks each instantiated one of the hypothesized line and the dot with a nonmagnetic fiber-optic track- developmental changes, but they were otherwise iden- ball (Current Designs, Philadelphia, USA). Performance tical to the ‘‘adult’’ network. was measured as the distance between the dot and the response location. Distracter trials were identical to vsWM trials, except for the additional presentation of Subjects and vsWM Task three yellow dots for 0.15 sec during the delay. The To test the predictions made by the developmental distracters appeared after 3, 6, or 9 sec of the 12-sec Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/5/750/1756680/jocn.2007.19.5.750.pdf by guest on 18 May 2021 hypotheses expressed in the computational model, delay period. In control trials, designed to control for we measured brain activity with fMRI in a group of 13 brain activity related to motor and visual functions, a children (10 boys, 13.1 ± 0.5 years) and 11 adults line crossed the screen for 0.5 sec, followed by a delay (4 men, 23 ± 3 years) while they performed a vsWM of 12 sec. After the delay, a blue dot appeared and the task. Behavioral data were successfully collected from task was to click on this circle. For both vsWM and 10 of the children and 9 of the adults. Significant ef- control trials, the maximum response time was 6 sec. fects of task and group, and the interaction between these factors, were calculated using a repeated-measures Brain Imaging: Analysis of Mean BOLD Signal analysis of variance (ANOVA). To get more reliable results for the interaction effects, additional data were in- The generalized linear model of fMRI time-series was cluded in the ANOVA. These data were collected outside applied using SPM2 (www.fil.ion.ucl.ac.uk/spm/software/ the scanner from 18 adults (8 men, 23.7 ± 1.9 years) spm2/) to statistically analyze the developmental changes and 9 children (6 boys, 12 ± 0.9 years). Thus, the main in mean BOLD signal (Friston et al., 1995). Hemodynam- performance analysis was based on data from 27 adults ic response functions in children and adults have previ- (11 men, 23.3 ± 2.4 years) and 19 children (13 boys, ously been shown to be comparable (Kang, Burgund, 12.7 ± 0.9 years).The children were recruited from a Lugar, Petersen, & Schlaggar, 2003). On the other hand, school in Solna, Sweden. The adults were recruited Thomason, Burrows, Gabrieli, and Glover (2005) re- through an advertisement on the hospital Web site ported that children had higher variability in the BOLD and through friends. All subjects were healthy and right signal. To test the possibility of unequal variances, we handed. Written consent was obtained from all subjects performed a two-tailed F test with significance level and from the parents of the children. The study was ap- a = .05 after correction for multiple comparisons (crit- proved by the ethical committee at Karolinska Institutet. ical values Flow = 0.930 and Fhi = 1.076) on the dataset The vsWM task (Figure 3) was adapted with the used for the correlation analysis (see the next section E-prime software (Psychology Software Tools, Pitts- for details about this dataset): Fpfc = 0.942, Fppc = 1.026. burgh, USA) from a previous study where it was used Effective degrees of freedom dfold = 3762, dfyoung = to successfully isolate vsWM maintenance-specific activ- 3764. The F ratios were within the critical values, and ity in the SFS and IPS (Rowe et al., 2000). An easy task thus, were not significant. was chosen so that both children and adults would In addition to the region-of-interest analysis of BOLD perform at a high level. All trials started with a fixation activity in the superior frontal and intraparietal regions cross and a 3-sec intertrial interval. For vsWM trials, included in the model, we also used an exploratory three blue dots appeared for 0.5 sec, followed by a delay analysis to investigate differences in delay-related brain phase of 12 sec, during which items were maintained activity in every voxel of the frontal and parietal corti- in vsWM. After the delay, a line crossed the location of ces. Clusters and voxels were considered as significant if p < .05, corrected for multiple comparisons. Correc- tions for multiple comparisons were based on the theory of Gaussian random fields (Worsley et al., 1996). Coor- dinates for localization of the activations were displayed in the MNI 152 space. For all statistical analyses of extracted voxel data, values that were more than 2 SD from the mean were excluded. Fixed-effect analyses were performed to create single-subject contrast images (Friston et al., 1998) of delay activity. Brain activity re- Figure 3. The vsWM task used in scanning and behavioral lated to the delay during the vsWM trial was compared experiments. Subjects fixated on the X. After an initial intertrial interval, to brain activity during the delay in the control trials. a cue consisting of three blue dots was presented, which subjects For each subject, one image was created for the delay were instructed to maintain in memory during the 12-sec-long delay period. At selection, a line appeared which crossed the position of contrast. To allow inferences to the population, random- one of the previously presented cue dots. At response, subjects had effects analyses were applied to the contrast images to click at that position. from the single-subject analyses. The main effect analy-

754 Journal of Cognitive Neuroscience Volume 19, Number 5 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.2007.19.5.750 by guest on 30 September 2021 ses of activity related to the delay-phase (i.e., the mean from 25 sessions from nine children were included in BOLD signal) consisted of one-sample t tests applied on the analysis. A fixed-effect analysis was performed by the linear combination of parameter estimates stored pooling BOLD signal values to calculate a single corre- in the contrast images. Interactions between group and lation coefficient for every group. The degrees of free- event-related activity were analyzed using a two-sample dom were calculated as described above in order to take t test on the contrast images. the autocorrelation between time points into account. The Fisher-transformed z-score was then calculated, and a one-sided normal test was performed to test for Computational Model and Brain Imaging: significant group difference in z-score. We used a one- Analysis of Fronto-parietal BOLD Correlations Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/5/750/1756680/jocn.2007.19.5.750.pdf by guest on 18 May 2021 sided test because the developmental effects of the Pearson’s product–moment coefficient of correlation correlation coefficient in the model were either positive between frontal and parietal BOLD signals was calcu- or not significant. lated with MATLAB 6.5 (The Mathworks, Natick, USA). The coefficient was transformed into an approximately 2 normally distributed Fisher z-score with variance sz = RESULTS (df2)1, where df is the effective degrees of freedom (Neter, Kutner, Nachtsheim, & Wasserman, 1996). Be- Increased Mean BOLD Signal Supports H1–H3 cause the slow dynamics of the hemodynamic response In the third column of Figure 2A, the simulated BOLD function introduces autocorrelations between BOLD curves of all five ‘‘child’’ networks (black) are compared signal measurement values at nearby time points, the to the ‘‘adult’’ network (green). Of the five hypotheses— correlation coefficient is measured with less certainty which were described in the Methods section—stronger than if measurements had been independent. Thispffiffiffiffiffiffiffiffiffiffiffiffi leads synaptic connections (H1, H2) and increased contrast to a reduction in df, from n 1 to (n r)/ 2pw2 (H3) resulted in a developmental increase of the BOLD (Friston et al., 1995). Here, n is the number of BOLD signal, whereas increased specificity (H4) caused a de- signal values and w = 1.42 is the smoothing factor, crease in the BOLD signal, and an increased conduction which is the standard deviation of a Gaussian curve velocity (H5) had no effect. fitted to the hemodynamic response, 2.84 sec, divided To assess the robustness of the simulated develop- by the time between measurements, 2 sec. Lastly, r is the mental effects on the mean BOLD signal, we developed number of effects modeled. In this case, r = 1 (the mean five additional versions of the computational model with BOLD signal). perturbed parameter values of the ‘‘child’’ and ‘‘adult’’ For the simulated hypotheses, four simulations were networks (Supplemental Tables 1 and 2). We changed made for each of the ‘‘adult’’ and ‘‘child’’ networks cor- synaptic locations, reduced the pyramidal cell model to responding to a developmental hypothesis, and t tests a single-compartment model, and removed all ion chan- were performed on the Fisher-transformed correlations nels except for the spike-producing Na+ and K+ chan- to test whether there were significant differences be- nels. In addition, we changed the absolute height and tween ‘‘child’’ and ‘‘adult’’ networks. width of the ‘‘adult’’ connection curve. In summary, To obtain correlations between experimentally mea- despite the considerable parameter changes, all tested sured BOLD signals, time-series of BOLD data were ex- parameter configurations produced the same relative dif- tracted from the SFS and IPS from each subject. In order ferences between ‘‘child’’ and ‘‘adult’’ mean BOLD sig- to maximize the task-related signal and minimize noise, nal as those shown in the third column of Figure 2A. we allowed the anatomical location where data were Furthermore, we confirmed that using only excitatory extracted to vary slightly from subject to subject so that synaptic currents or total spiking activity, instead of we could find the voxel with maximum task-related total synaptic currents, as a basis for the BOLD signal activity within the SFS and IPS in each subject. If no (Mukamel et al., 2005; Attwell & Iadecola, 2002) did not task-related activity was found within the region for a affect our results. particular subject and session (using a liberal threshold We next compared the predicted BOLD activity to the of p < .3 and cluster size > 90), no data were extracted delay-phase BOLD activity measured with fMRI. Delay from that session. The application of the liberal threshold activity was extracted from maxima in the SFS and IPS served to further enhance signal-to-noise ratio. An equal in both hemispheres (Figure 2B, column 2). Although number of sessions were excluded from each group. previous imaging studies have been performed on From the identified maximum voxels, we defined one the development of vsWM-related BOLD activity, the parietal and one frontal region of interest with 3 mm delay-phase activity has not been previously isolated. radii, extracted the time-series of BOLD values from We therefore conducted an experiment to measure this within this VOI, and used the eigenvector of that data activity in SFS and IPS. A more comprehensive analysis for further analysis. Altogether, a total of 5323 BOLD of all imaging results relating to the development of signal values from 25 sessions (with 213 time points per vsWM from this experiment can be found in Olesen session) from nine adults and 5326 BOLD signal values et al. (2006). An ANOVA with the behavioral data showed

Edin et al. 755 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.2007.19.5.750 by guest on 30 September 2021 signal were predicted by H1 and H2: stronger intra- and interregional synaptic connections, and H3: increased contrast, but not by H4: increased specificity or H5: increased conduction velocity.

Increased Interregional Correlation of BOLD Signal Supports H2–H3 H1, H2, and H3 were the only hypotheses that were con- Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/5/750/1756680/jocn.2007.19.5.750.pdf by guest on 18 May 2021 sistent with the developmental increase in mean BOLD signal. To distinguish between these three hypotheses, we analyzed interregional correlations, based on the idea that brain activity in regions that are strongly connected should be correlated. By performing repeated simulations, we could statistically confirm that the ‘‘adult’’ network had a higher interregional correlation of the BOLD signal than the ‘‘child’’ network for both H2 and H3 (difference in Fisher z-score, Á, was 0.59, p < .001, n =8 for H2 and Á = 0.48, p < .05, n = 8 for H3). Changing local connection strength (H1, p = .23, n =8)orin- creasing neuronal conduction velocity (H5, p = .15, n = 8) produced no difference, whereas increasing specificity (H4, Á = 0.35, p < .05, n = 8) led to a decreased Figure 4. vsWM performance and distractibility in experiments and simulations. (A) Perturbing currents were injected into the interregional correlation. We then calculated group dif- somata of cells located 1808 away from the cue (arrows). Frontal ferences in correlations from the measured BOLD time- region pyramidal cell activity is shown. Each dot represents an series data. We found that adults had a significantly action potential. (B) Adults were significantly more accurate than higher correlation coefficient, r, between frontal and children on the basic WM task (‘‘No Distraction’’). In separate trials parietal activities (r = 0.42, r = 0.37 and Á = (‘‘Distraction’’), distracting dots were presented during the delay adult child period. The distracter impaired performance significantly more for 0.06, p < .05). This confirmed the prediction made by children than for adults. Performance was measured as the error in H2: stronger interregional synaptic connections, and H3: millimeters between cue and response locations. Error bars: SEM. increased contrast, but did not confirm H1: stronger Green = adult; black = child. intraregional synaptic connections (or H4: increased specificity and H5: increased conduction velocity). that the adults were significantly more accurate on the vsWM task ( p < .001; Figure 4B, ‘‘No distraction’’), H2–H3 Predict Human Distractibility confirming previous findings (Gathercole et al., 2004). An additional two-way ANOVA with fMRI data showed Lastly, to assess the predictive capacity of the H2–H3 greater activity in adults than in children [F(1, 44) = network models, we investigated their mnemonic per- 4.52, p < .05], but found no interaction between age and formance by testing whether memory-related activity region [F(1, 44) = 0.89, p = .35], which is consistent was affected by a perturbing current (Figure 4A). Activ- with the presupposed symmetric structure of the com- ities in the ‘‘child’’ networks H2 and H3 (and H1) were putational model. Figure 2B, column 3, shows the mea- less stable than the ‘‘adult’’ activity (whereas H4 and H5 sured delay activity, averaged across bilateral SFS and IPS ‘‘child’’ networks were not), which is consistent with the regions, for the adult and child groups. finding that adult subjects show a greater resistance to We also performed an exploratory analysis using SPM2 distracters (Figure 4B). This result serves as an indepen- to investigate differences in delay-related brain activity in dent validation of the results from the investigations of every voxel of the frontal and parietal cortices. We found mean BOLD signal and interregional BOLD signal corre- that adults showed higher activity than children in the lations as discussed previously. right middle frontal gyrus and intraparietal cortex, whereas there were no regions in which children showed Conclusion—Synaptic Strengthening behind higher BOLD activity than adults. The location of this Developmental vsWM Improvement frontal region corresponds to an anterior frontal activation, presumably Brodmann’s area 46. This indicates that To conclude, we found that three types of changes— the relative difference between adult and child delay increased intraregional connectivity (H1), increased activity does not depend on the particular location of interregional connectivity (H2), and increased contrast measurements in the frontal and parietal lobe. In conclu- (H3)—could predict the developmental increase in delay- sion, the results from measurements of the mean BOLD phase BOLD activity in the frontal and parietal lobes, as

756 Journal of Cognitive Neuroscience Volume 19, Number 5 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.2007.19.5.750 by guest on 30 September 2021 well as the superior performance of adults on the vsWM et al. (2001) reported a correlation coefficient of .72 be- task. Of these hypotheses, H2 and H3 also predicted the tween the local field potential and the BOLD signal, and developmental increase in the interregional correlation of .67 between multiunit activity and the BOLD signal. of activity. Thus, H2 and H3 were the only mechanisms Mukamel et al. (2005) showed that a convolution be- that were supported by the experimental data. tween either spiking activity or local field potential and Although slightly different, in both H2 and H3, syn- a linear hemodynamic response function was a good aptic strengthening causes an increase in the interre- predictor of BOLD activity. By excluding time periods gional connection strength between cells coding for when there was no neural activity in the brain area un- similar stimuli. The impact of a change in strength of a der investigation—such periods, by definition, result in Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/5/750/1756680/jocn.2007.19.5.750.pdf by guest on 18 May 2021 connection going from cell i to cell j is high if cell i has zero correlation without implying poor predictability— high activity, whereas it is low if the activity in cell i is they were able to show even higher predictive power low. Therefore, synaptic strengthening in H2 and H3 has of the spiking activity, with a correlation coefficient of a major impact on vsWM-related activity, as it causes an .81, and reaching .9 during the most informative time increased connection strength between those cells in periods. the network having the highest activity (those coding With regard to the biological plausibility of the cor- for the stimulus). By the same token, synaptic pruning tical model, it should be noted that an important as- has a low effect on mnemonic activity because the de- sumption of the model is that the two regions only crease in synaptic strength occurs in connections from interact through their pyramidal cells. This assumption cells with low activity. We therefore conclude that it is most likely is an oversimplification, as excitatory cells the strengthening of fronto-parietal synapses, rather could make long-range connections to inhibitory cells. than synaptic pruning, that constitutes the most impor- However, a study of the effect of cooling either the pre- tant cellular maturational process underlying the im- frontal cortex or the posterior parietal cortex on the provement of vsWM. activity of the other region suggests that net connections are excitatory (Chafee & Goldman-Rakic, 2000). There- fore, a nonspecific strengthening of connections onto excitatory and inhibitory cells alike should generate the DISCUSSION same change in mean BOLD signal as was seen in the In the present study, we have used a new approach in present study. Similarly, the interregional correlations order to evaluate how developmental processes taking between BOLD signals are not expected to be affected place on the neuronal level can be directly related to by the omission of interregional excitatory connections the maturation of WM and associated changes in brain to inhibitory cells, because it is still the case that activity activity. Deco et al. (2004) have developed a similar ap- in strongly connected regions will be correlated. proach comparing modeling and fMRI, but their ap- The frontal and parietal regions identified in this study proach is slightly different in that they propose new were previously shown to be active during the delay models which can account for a previously observed period in vsWM tasks (Curtis et al., 2004; Rowe et al., phenomenon, whereas we use modeling to make pre- 2000; Courtney et al., 1998). Other fMRI studies of vsWM dictions about unknown phenomena (the BOLD signal comparing children and adults (Bunge et al., 2002; characteristics had never been measured before). Thus, Klingberg et al., 2002; Casey et al., 2000) were also com- by expressing neuronal developmental changes in a patible with our data, although none of these isolated computational model that represents our current knowl- the delay-specific activity simulated in the model. Espe- edge about the mechanisms underlying the delay-phase cially noteworthy is the increased fronto-parietal con- activity studied in vsWM tasks, we could predict the nection strength found in this study which is compatible effect of these changes on the magnitude and corre- with a previous study showing a higher correlation be- lation of delay-phase activity in two vsWM-related brain tween area 8 and the posterior part of the IPS in correct regions. We found that one developmental change ac- than incorrect trials (Sakai, Rowe, & Passingham, 2002). counted for the observed increase in vsWM-related brain Apart from the developmental processes dealt with in activity, namely, synaptic strengthening, which results in this study, it is possible that other processes—such as the increased synaptic connection strength between cells in development of the dopamine system—could improve SFS and IPS with similar coding preferences. Other po- WM as well. The strength of connection between an tential maturational processes, such as pruning and upstream and a downstream neuron is defined as the myelination, could not account for the observed devel- influence that a unit change in the activity of the up- opmental changes in brain activity. stream neuron has on the downstream neuron. Although The conclusion reached in this study could be affect- maturation of synapses is the most natural mechanism ed either by discrepancies between the model and the whereby interregional connections could be strength- human brain or by an incomplete model of the neuro- ened, there are also other ways to achieve this. Modu- vascular coupling function. The feasibility of the latter latory neurotransmitters such as dopamine might act by has been supported in two recent articles. Logothetis changing effective connection strengths between active

Edin et al. 757 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.2007.19.5.750 by guest on 30 September 2021 cells. Dopamine levels are increased during WM tasks We do not rule out that several of the mechanisms (Watanabe, Kodama, & Hikosaka, 1997), and blocking described above could interact during development. In of the dopamine D1 receptor can increase the signal- fact, building on the work in the present study, a recent to-noise ratio of the delay-phase activity, that is, the study by Macoveanu et al. (2006) shows that although activity of delay-phase pyramidal neurons tuned to the specificity on its own does not improve vsWM capacity, angle of the stimulus increases, whereas the activity of specificity and increased mean connection strength or the neurons with other preferred directions decreases increased contrast have a synergistic effect on capacity. (Williams & Goldman-Rakic, 1995). Moreover, dopa- However, in separate simulations, we have seen no in- mine innervation of the prefrontal cortex declines dur- teraction between conduction velocity and interregional Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/5/750/1756680/jocn.2007.19.5.750.pdf by guest on 18 May 2021 ing adolescence in the primate (Rosenberg & Lewis, connection strength on interregional synchrony or mean 1995), suggesting a possible functional increase in the activity levels (data not shown). It is possible that the contrast (H3) of interregional connections through a combination of increased specificity and stronger con- developmental change in D1 stimulation. However, sev- nections or contrast could lead to stronger but more eral effects of dopamine on the properties of prefrontal focused activity during development, as has been ob- layer III pyramidal cells have been reported (Gonzalez- served during cognitive development other than vsWM Burgos et al., 2002), and their interrelationship and (Casey, Galvan, & Hare, 2005). However, we did not test relative importance is still unknown. Further research this explicitly in this study. is therefore needed before any reliable predictions can Lastly, it is possible that the mechanisms we have be made on how dopamine modulation of WM-related identified as important for development are more gen- brain activity changes during development. erally valid for determining interindividual differences in Our modeling did not suggest any effect of myelin on WM capacity. The link between delay-related activity in brain activity levels. This might seem inconsistent with the intraparietal cortex and vsWM capacity that we found previous studies suggesting the importance of myelina- by studying developmental changes has also previously tion in general (Giedd et al., 1999; Klingberg et al., 1999; been identified within the context of interindividual var- Paus et al., 1999; Sowell et al., 1999) and, more specif- iability in vsWM capacity among adult subjects (Olesen, ically, a connection between fronto-parietal myelination Westerberg, & Klingberg, 2004; Todd & Marois, 2004). and WM maturation (Nagy, Westerberg, & Klingberg, 2004; Olesen et al., 2003). However, from a dynamical Acknowledgments system’s point of view, it is not surprising that changes in conduction velocity did not affect the activity during This work was supported by the Foundation for Strategic the delay period. The mnemonic activity is a stable state Research and the Wallenberg Global Learning Network. Fredrik Edin was supported by a PhD Position of Excellence at the Royal in the network (a fixpoint of the system). Although the Institute of Technology, Stockholm, Sweden. Terkel Klingberg change in conduction velocity could possibly have was supported by a research position from the Royal Academy caused the mnemonic activity to synchronize (oscillate of Sciences through funding from the Knut and Alice Wallenberg around the fixpoint, a known possible effect of incorpo- Foundation. We thank Albert Compte, Erik Franse´n, and Anders rating conduction delays into dynamical systems), which Ledberg for valuable comments on the manuscript. The idea was conceived by Jesper Tegne´r and Torkel Klingberg. Fredrik could, in turn, have affected the mean neuronal firing Edin implemented the model and carried out the simulations rate in the network, this was not seen (data not shown). and model analysis. Julian Macoveanu, Pernille J. Olesen, and Indeed, the most natural assumption would be that be- Torkel Klingberg performed imaging experiments. Fredrik Edin, cause the network has reached a stable state—meaning Jesper Tegne´r, and Torkel Klingberg wrote the article. that the mean activity level in the network is approxi- Reprint requests should be sent to Torkel Klingberg, Pediatric mately the same between reasonably proximal time Neurology, Department of Woman and Child Health MR Cen- points—the incoming synaptic activity experienced by ter, Karolinska Institutet, SE-171 76 Stockholm, Sweden, or via a neuron would be approximately the same regardless e-mail: [email protected]; or to Jesper Tegne´r, Compu- tational Biology, Department of Physics, Linko¨ping University of whether this activity originated 6 or 12 msec before. It of Technology, SE-581 53 Linko¨ping, Sweden, or via e-mail: is, however, possible that myelination could cause a de- [email protected]. creased action potential failure rate (Zhou & Chiu, 2001), and the effect of myelination could thus be to increase the influence that a change in the activity of the REFERENCES upstream neuron has on the downstream neuron, that Amit, D. J., & Brunel, N. (1997). Model of global spontaneous is, an increase in connection strength. Finally, we must activity and local structured activity during delay periods stress that the absence of an effect of conduction veloc- in the cerebral cortex. Cerebral Cortex, 7, 237–252. ity on delay-phase activity does not mean that myelina- Attwell, D., & Iadecola, C. (2002). The neural basis of tion is unimportant for vsWM. Indeed, conduction functional brain imaging signals. Trends in Neurosciences, 25, 621–625. velocity might be very important for other processes Baddeley, A. D., & Hitch, G. J. (1974). Working memory. related to vsWM rather than maintenance, for example, In G. H. Bower (Ed.), Recent advances in learning and encoding. motivation (pp. 47–49). New York: Academic Press.

758 Journal of Cognitive Neuroscience Volume 19, Number 5 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.2007.19.5.750 by guest on 30 September 2021 Bourgeois, J. P., Goldman-Rakic, P. S., & Rakic, P. (2000). Gathercole, S. E., Pickering, S. J., Ambridge, B., & Wearing, H. Formation, elimination, and stabilization of synapses in (2004). The structure of working memory from 4 to the primate cerebral cortex. In M. S. Gazzaniga (Ed.), The 15 years of age. Developmental Psychology, 40, 177–190. new cognitive neurosciences (pp. 45–53). Cambridge: Giedd, J. N., Blumenthal, J., Jeffries, N. O., Castellanos, F. X., Massachusetts Institute of Technology. Liu, H., Zijdenbos, A., et al. (1999). Brain development Bunge, S. A., Dudukovic, N. M., Thomason, M. E., Vaidya, during childhood and adolescence: A longitudinal MRI C. J., & Gabrieli, J. D. (2002). Immature frontal lobe study. Nature Neuroscience, 2, 861–863. contributions to cognitive control in children: Evidence Gonzalez-Burgos, G., Kroner, S., Krimer, L. S., Seamans, from fMRI. Neuron, 33, 1–11. J. K., Urban, N. N., Henze, D. A., et al. (2002). Dopamine Casey, B. J., Galvan, A., & Hare, T. A. (2005). Changes modulation of neuronal function in the monkey prefrontal

in cerebral functional organization during cognitive cortex. Physiology & Behavior, 77, 537–543. Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/5/750/1756680/jocn.2007.19.5.750.pdf by guest on 18 May 2021 development. Current Opinion in Neurobiology, 15, Hubel, D. H., & Wiesel, T. N. (1963). Receptive fields of 239–244. cells in striate cortex of very young, visually inexperienced Casey, B. J., Giedd, J. N., & Thomas, K. M. (2000). kittens. Journal of Neurophysiology, 26, 994–1002. Structural and functional brain development and its Huttenlocher, P. R. (1979). Synaptic density in human frontal relation to cognitive development. Biological Psychology, cortex—Developmental changes and effects of aging. 54, 241–257. Brain Research, 163, 195–205. Chafee, M. V., & Goldman-Rakic, P. S. (2000). Inactivation Kang, H. C., Burgund, E. D., Lugar, H. M., Petersen, S. E., of parietal and prefrontal cortex reveals interdependence & Schlaggar, B. L. (2003). Comparison of functional of neural activity during memory-guided saccades. Journal activation foci in children and adults using a common of Neurophysiology, 83, 1550–1566. stereotactic space. Neuroimage, 19, 16–28. Chafee, M. V., & Goldman-Rakic, P. S. (1998). Matching Klingberg, T., Forssberg, H., & Westerberg, H. (2002). patterns of activity in primate prefrontal area 8a and Increased brain activity in frontal and parietal cortex parietal area 7ip neurons during spatial working memory. underlies the development of visuospatial working Journal of Neurophysiology, 79, 2919–2940. memory capacity during childhood. Journal of Cognitive Compte, A. (2006). Computational and in vitro studies of Neuroscience, 14, 1–10. persistent activity: Edging towards cellular and synaptic Klingberg, T., Vaidya, C. J., Gabrieli, J. D., Moseley, M. E., & mechanisms of working memory. Neuroscience, 139, Hedehus, M. (1999). Myelination and organization of the 135–151. frontal white matter in children: A diffusion tensor MRI Compte, A., Brunel, N., Goldman-Rakic, P. S., & Wang, X. J. study. NeuroReport, 10, 2817–2821. (2000). Synaptic mechanisms and network dynamics Lamantia, A. S., & Rakic, P. (1990). Cytological and quantitative underlying spatial working memory in a cortical network characteristics of four cerebral commissures in the rhesus model. Cerebral Cortex, 10, 910–923. monkey. Journal of Comparative Neurology, 291, 520–537. Constantinidis, C., Franowicz, M. N., & Goldman-Rakic, Logothetis, N. K., Pauls, J., Augath, M., Trinath, T., & P. S. (2001). Coding specificity in cortical microcircuits: Oeltermann, A. (2001). Neurophysiological investigation A multiple-electrode analysis of primate prefrontal cortex. of the basis of the fMRI signal. Nature, 412, 150–157. Journal of Neuroscience, 21, 3646–3655. Macoveanu, J., Klingberg, T., & Tegne´r, J. (2006). A biophysical Courtney, S. M., Petit, L., Maisog, J. M., Ungerleider, L. G., & model of multiple-item working memory: A computational Haxby, J. V. (1998). An area specialized for spatial working and neuroimaging study. Neuroscience, 141, 1611–1618. memory in human frontal cortex. Science, 279, 1347–1351. Mukamel, R., Gelbard, H., Arieli, A., Hasson, U., Fried, I., Curtis, C. E., Rao, V. Y., & D’Esposito, M. (2004). Maintenance & Malach, R. (2005). Coupling between neuronal firing, of spatial and motor codes during oculomotor delayed field potentials, and fMRI in human auditory cortex. response tasks. Journal of Neuroscience, 24, 3944–3952. Science, 309, 951–954. Deco, G., Rolls, E. T., & Horwitz, B. (2004). ‘‘What’’ and Nagy, Z., Westerberg, H., & Klingberg, T. (2004). Maturation ‘‘where’’ in visual working memory: A computational of white matter is associated with the development of neurodynamical perspective for integrating fMRI and cognitive functions during childhood. Journal of Cognitive single-neuron data. Journal of Cognitive Neuroscience, Neuroscience, 16, 1227–1233. 16, 683–701. Neter, J., Kutner, M. H., Nachtsheim, C. J., & Wasserman, W. Douglas, R. J., & Martin, K. A. (2004). Neuronal circuits of (1996). Applied linear statistical models (4th ed.). the neocortex. Annual Review of Neuroscience, 27, Chicago: Irwin. 419–451. Olesen, P. J., Macoveanu, J., Tegne´r, J., & Klingberg, T. (2006). Ferraina, S., Pare, M., & Wurtz, R. H. (2002). Comparison Brain activity related to working memory and distraction in of cortico-cortical and cortico-collicular signals for the children and adults. Cerebral Cortex, [Epub ahead of print]. generation of saccadic eye movements. Journal of Olesen, P. J., Nagy, Z., Westerberg, H., & Klingberg, T. (2003). Neurophysiology, 87, 845–858. Combined analysis of DTI and fMRI data reveals a joint Friston, K. J., Fletcher, P., Josephs, O., Holmes, A., Rugg, M. D., maturation of white and grey matter in a fronto-parietal & Turner, R. (1998). Event-related fMRI: Characterizing network. Brain Research, Cognitive Brain Research, 18, differential responses. Neuroimage, 7, 30–40. 48–57. Friston, K. J., Holmes, A. P., Poline, J. B., Grasby, P. J., Williams, Olesen, P. J., Westerberg, H., & Klingberg, T. (2004). Increased S. C., Frackowiak, R. S., et al. (1995). Analysis of fMRI prefrontal and parietal activity after training of working time-series revisited. Neuroimage, 2, 45–53. memory. Nature Neuroscience, 7, 75–79. Fry, A. F., & Hale, S. (2000). Relationships among processing Paus, T., Zijdenbos, A., Worsley, K., Collins, D. L., Blumenthal, speed, working memory, and fluid intelligence in children. J., Giedd, J. N., et al. (1999). Structural maturation of Biological Psychology, 54, 1–34. neural pathways in children and adolescents: In vivo Funahashi, S., Bruce, C. J., & Goldman-Rakic, P. S. (1989). study. Science, 283, 1908–1911. Mnemonic coding of visual space in the monkey’s Rainer, G., & Miller, E. K. (2000). Effects of visual experience dorsolateral prefrontal cortex. Journal of Neurophysiology, on the representation of objects in the prefrontal cortex. 61, 331–349. Neuron, 27, 179–189.

Edin et al. 759 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.2007.19.5.750 by guest on 30 September 2021 Rakic, P., Bourgeois, J. P., Eckenhoff, M. F., Zecevic, N., & mnemonic persistent activity. Trends in Neurosciences, Goldman-Rakic, P. S. (1986). Concurrent overproduction 24, 455–463. of synapses in diverse regions of the primate cerebral Wang, X. J., Tegne´r, J., Constantinidis, C., & Goldman-Rakic, cortex. Science, 232, 232–235. P. S. (2004). Division of labor among distinct subtypes Rosenberg, D. R., & Lewis, D. A. (1995). Postnatal of inhibitory neurons in a cortical microcircuit of maturation of the dopaminergic innervation of monkey working memory. Proceedings of the National Academy prefrontal and motor cortices: A tyrosine hydroxylase of Sciences, U.S.A., 101, 1368–1373. immunohistochemical analysis. Journal of Comparative Watanabe, M., Kodama, T., & Hikosaka, K. (1997). Increase Neurology, 358, 383–400. of extracellular dopamine in primate prefrontal cortex Rowe, J. B., Toni, I., Josephs, O., Frackowiak, R. S., & during a working memory task. Journal of Neurophysiology,

Passingham, R. E. (2000). The prefrontal cortex: Response 78, 2795–2798. Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/5/750/1756680/jocn.2007.19.5.750.pdf by guest on 18 May 2021 selection or maintenance within working memory? Westerberg, H., Hirvikoski, T., Forssberg, H., & Klingberg, T. Science, 288, 1656–1660. (2004). Visuo-spatial working memory span: A sensitive Sakai, K., Rowe, J. B., & Passingham, R. E. (2002). Active measure of cognitive deficits in children with ADHD. maintenance in prefrontal area 46 creates distractor-resistant Child Neuropsychology, 10, 155–161. memory. Nature Neuroscience, 5, 479–484. Williams, G. V., & Goldman-Rakic, P. S. (1995). Modulation Sowell, E. R., Thompson, P. M., Holmes, C. J., Jernigan, T. L., of memory fields by dopamine D1 receptors in prefrontal & Toga, A. W. (1999). In vivo evidence for post-adolescent cortex. Nature, 376, 572–575. brain maturation in frontal and striatal regions. Nature Worsley, K., Marret, S., Neeling, P., Vandal, A. C., Friston, K., Neuroscience, 2, 859–861. & Evans, A. C. (1996). A unified statistical approach for Tegne´r, J., Compte, A., & Wang, X. J. (2002). The dynamical determining significant signals in images of cerebral stability of reverberatory neural circuits. Biological activation. Human Brain Mapping, 4, 58–73. Cybernetics, 87, 471–481. Yakovlev, P. I., & Lecours, A.-R. (1967). The myelogenetic Thomason, M. E., Burrows, B. E., Gabrieli, J. D., & Glover, G. H. cycles of regional maturation of the brain. In (2005). Breath holding reveals differences in fMRI BOLD A. Minkowski (Ed.), Regional development of the signal in children and adults. Neuroimage, 25, 824–837. brain in early life (pp. 3–65). Oxford: Blackwell Todd, J. J., & Marois, R. (2004). Capacity limit of visual Scientific Publications. short-term memory in human posterior parietal cortex. Zhou, L., & Chiu, S. Y. (2001). Computer model for action Nature, 428, 751–754. potential propagation through branch point in myelinated Wang, X. J. (2001). Synaptic reverberation underlying nerves. Journal of Neurophysiology, 85, 197–210.

760 Journal of Cognitive Neuroscience Volume 19, Number 5 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.2007.19.5.750 by guest on 30 September 2021